Leveraging Behavioral Patterns of Mobile Applications for …yvchen/doc/ICMI15_Multi... ·...

Leveraging Behavioral Patterns of Mobile Applications for

Personalized Spoken Language Understanding

YUN-NUNG (VIVIAN) CHENMING SUN

ALEXANDER I RUDNICKYANATOLE GERSHMAN

Introduction• Global Digital Statistics (2015 Jan)

Global Population

7.21B

Active Internet Users

3.01B

Active Social Accounts

2.08B

Active Mobile Users

3.65B

Spoken language interfaces become important are incorporated in smart devices as intelligent assistants (IAs).

Spoken language understanding (SLU) is a key component of IA, which predicts users’ intended apps by understanding input utterances.

2

Introduction

• Task: user intent prediction

• Challenge: language ambiguity

User preference

Some people prefer “Message” to “Email”

Some people prefer “Outlook” to “Gmail”

App-level contexts

“Message” is more likely to follow “Camera”

“Email” is more likely to follow “Excel”

send to vivianv.s.

Email? Message?

Communication

Considering behavioral patterns in history to model SLU for intent prediction.3

Data Collection

• Subjects’ app invocation is logged on a daily basis

• Subjects annotate their app activities with

– Task Structure: link applications that serve a common goal

– Task Description: briefly describe the goal or intention of the task

• Subjects use a wizard system to perform the annotated task by speech

TASK59; 20150203; 1; Tuesday; 10:48

play music via bluetooth speaker

com.android.settings com.lge.music

Meta

Desc

App

: Ready.

: Connect my phone to bluetooth speaker.

: Connected to bluetooth speaker.

: And play music.

: What music would you like to play?

: Shuffle playlist.

: I will play the music for you.

W1

U1

W2

U2

W3

U3

W4

DialogueSETTINGS

MUSIC

MUSIC

4

SLU for Intent Prediction

1

Lexical Intended App

photo check CAMERA IMtell

take this phototell vivian this is me in the lab

CAMERA

IM

Traincheck my grades on websitesend an email to professor

…

CHROME

EMAIL

send

Behavioral

NULL CAMERA

.85

take a photo of thissend it to alice

CAMERA

IM

…

email

1

1

1 1

1

1 .70

CHROME

1

1

1

1

1

1

CHROME EMAIL

1

1

1

1

.95

.80 .55

User UtteranceIntended

App

Testtake a photo of thissend it to alex

…

hidden semantics

Issue: unobserved hidden semantics may benefit understanding

Solution: use matrix factorization to complete a partially-missing matrix based on a low-rank latent semantics assumption.

5

Matrix Factorization (MF)

1

Lexical Intended App

photo check CAMERA IMtell send

Behavioral

NULL CAMERA

.85

email

1

1

1 1

1

1 .70

CHROME

1

1

1

1

1

1

CHROME EMAIL

1

1

1

1

.95

.80 .55

𝑼

𝑾 + 𝑯 + 𝑨

≈ 𝑼 × 𝒅 𝒅 × 𝑾 + 𝑯 + 𝑨

• The decomposed matrices represent low-rank latent semantics for

utterances and words/histories/apps respectively

• The product of two matrices fills the probability of hidden semantics

6

Parameter Estimation

• Model implicit feedback by completing the matrix

– not treat unobserved facts as negative samples (true or false)

– give observed facts higher scores than unobserved facts

• Objective:

– the model can be achieved by SGD updates with fact pairs

1

𝑓+ 𝑓− 𝑓−

𝑢

𝑥

The objective is to learn a set of well-ranked apps per utterance.

7

SLU Modeling by MF

1

Lexical Intended App

photo check CAMERA IMtell

take this phototell vivian this is me in the lab

CAMERA

IM

Traincheck my grades on websitesend an email to professor

…

CHROME

EMAIL

send

Behavioral

NULL CAMERA

.85

take a photo of thissend it to alice

CAMERA

IM

…

email

1

1

1 1

1

1 .70

CHROME

1

1

1

1

1

1

CHROME EMAIL

1

1

1

1

.95

.80 .55

User UtteranceIntended

App

Reasoning with Matrix Factorization for Implicit Intents

Testtake a photo of thissend it to alex

…

8

Experiments

Approach Lexical Behavioral All

(a)MLE

User-Indep 13.5 / 19.6

(b) User-Dep 20.2 / 27.9

• Dataset: 533 dialogues (1,607 utterances); 455 multi-turn dialogues

• Google recognized transcripts (word error rate = 25%)

• Evaluation metric: accuracy of user intent prediction (ACC)mean average precision of ranked intents (MAP)

• Baseline: Maximum Likelihood Estimation (MLE)Multinomial Logistic Regression (MLR)

9

The user-dependent model is better than the user-independent model.

Experiments

Approach Lexical Behavioral All

(a)MLE

User-Indep 13.5 / 19.6

(b) User-Dep 20.2 / 27.9

(c)MLR

User-Indep 42.8 / 46.4 14.9 / 18.7

(d) User-Dep 48.2 / 52.1 19.3 / 25.2

• Dataset: 533 dialogues (1,607 utterances); 455 multi-turn dialogues

• Google recognized transcripts (word error rate = 25%)

• Evaluation metric: accuracy of user intent prediction (ACC)mean average precision of ranked intents (MAP)

• Baseline: Maximum Likelihood Estimation (MLE)Multinomial Logistic Regression (MLR)

Lexical features are useful to predict intended apps for both user-independent and user-dependent models.

10

Experiments

Approach Lexical Behavioral All

(a)MLE

User-Indep 13.5 / 19.6

(b) User-Dep 20.2 / 27.9

(c)MLR

User-Indep 42.8 / 46.4 14.9 / 18.7 46.2+ / 50.1+

(d) User-Dep 48.2 / 52.1 19.3 / 25.2 50.1+ / 53.9+

• Dataset: 533 dialogues (1,607 utterances); 455 multi-turn dialogues

• Google recognized transcripts (word error rate = 25%)

• Evaluation metric: accuracy of user intent prediction (ACC)mean average precision of ranked intents (MAP)

• Baseline: Maximum Likelihood Estimation (MLE)Multinomial Logistic Regression (MLR)

Combining lexical and behavioral features improves performance of the MLR model, which models explicit information from observations.

11

Experiments

Approach Lexical Behavioral All

(a)MLE

User-Indep 13.5 / 19.6

(b) User-Dep 20.2 / 27.9

(c)MLR

User-Indep 42.8 / 46.4 14.9 / 18.7 46.2+ / 50.1+

(d) User-Dep 48.2 / 52.1 19.3 / 25.2 50.1+ / 53.9+

(e) (c) + Personalized MF 47.6 / 51.1 16.4 / 20.3 50.3+* / 54.2+*

(f) (d) + Personalized MF 48.3 / 52.7 20.6 / 26.7 51.9+* / 55.7+*

• Dataset: 533 dialogues (1,607 utterances); 455 multi-turn dialogues

• Google recognized transcripts (word error rate = 25%)

• Evaluation metric: accuracy of user intent prediction (ACC)mean average precision of ranked intents (MAP)

• Baseline: Maximum Likelihood Estimation (MLE)Multinomial Logistic Regression (MLR)

Personalized MF significantly improves MLR results by considering hidden semantics.12

Conclusion

• An MF model exploits both lexical and behavioral features for SLU,

which considers implicit semantics to enhance intent inference given

the noisy ASR inputs.

• We are able to model users’ contextual behaviors and their app

preference for better intent prediction.

• The proposed multi-model personalized SLU effectively improves

intent prediction performance, achieving about 52% on turn accuracy

and 56% on mean average precision for ASR transcripts with 25%

word error rate.

13

THANKS FOR ATTENTIONS!

Q&AData Available at

http://AppDialogue.com